Real-Time Food Quality Assessment of Oyster Mushrooms Using an Electronic Nose

India, known for its vast agricultural potential, has seen impressive growth in mushroom cultivation, especially oyster mushrooms (Pleurotus ostreatus). These mushrooms stand out for being affordable, nutritious, and increasingly sought after by consumers. Over the last decade, mushroom production has surged, meeting both local and export demands. However, the highly perishable nature of oyster mushrooms poses significant challenges in their post-harvest handling and storage. With moisture levels as high as 80-90. As a result, mushroom quality often deteriorates rapidly, leading to sensory and nutritional losses and substantial economic setbacks. 

Spoilage in oyster mushrooms is caused by both internal and external factors. The internal ones include high microbial loads post-harvest-between 2.5 and 5.8 log cfu/g - and enzymatic activity, which cause discoloration, softening, and the development of off-odors. External factors during transportation, such as high temperature, humidity, and poor packaging, also increase spoilage. Those challenges show the great need for innovative, non-destructive methods in monitoring and detecting spoilage signs at early stages. 

One is the use of an electronic nose (E-nose), which mimics the human sense of smell to detect Volatile Organic Compounds (VOCs) that mushrooms emit as they begin to spoil. Compounds like alcohols, ketones, and sulfur-based molecules act as clear indicators of microbial activity and metabolic changes. E-noses provide instant and non-invasive spoilage detection, preserving the quality of mushrooms while offering a practical monitoring tool. Linked with the IoT-enabled platforms, E-noses prove even more valuable by enabling real-time tracking and remote monitoring-ideally fitting for India’s complex supply chain needs. 

Success in the adoption of E-nose technology in India would require addressing the following factors: sensor arrays tuned for the recognition of spoilage-specific VOCs under local conditions, advanced data analysis through machine learning or other methods to classify stages of spoilage and predict shelf life, integration of IoT modules that allows seamless sharing of data via cloud platforms for easy access to actionable insights by farmers and distributors, and finally, user-friendly apps that can offer timely alerts and practical recommendations for stakeholders to minimize losses. 

E-nose technology can transform India’s mushroom industry by reducing post-harvest losses and improving food safety. Real-time monitoring could prevent quality degradation, ensuring that mushrooms retain their market value and reduce financial strain on farmers and suppliers. Combining E-nose systems with predictive models and IoT frameworks offers a scalable and cost-efficient solution for creating a sustainable agricultural supply chain. 

In conclusion, E-nose technology is a game-changer in the management of spoilage in oyster mushroom cultivation. Accurate, real-time, and non-destructive monitoring guarantees that quality produce is delivered consistently, with less wastage. The way forward should be in the fine-tuning of sensor arrays, enhancing IoT integration, and conducting field tests to validate the technology in real-world scenarios. This would go a long way in ensuring maximum impact and in supporting a vibrant mushroom farming sector in India.

Extensive researches found that the monitoring of spoilage and freshness in perishable foods using the technique of an electronic nose can help determine VOC. Wang et al. [1], considered alcohols, acids, and aldehydes to be the significant agents causing oyster mushroom spoilage by analyzing these substances in the products of GC-MS and E-nose. Their application of Linear Discriminant Analysis (LDA) with machine learning algorithms, SVM, and RF attained over 90% classification accuracy with different storage conditions, and therefore, has good potential in the E-nose system. Researched the sensory and microbiological quality of the stored mushrooms after cultivation. In that study, significant correlations between the spoilage-related sensory changes, microbial activity, and environmental factors such as temperature and humidity were found. Emphasized versatility in E-nose technology food freshness detection using sensor arrays with VOC classification abilities to simulate a human olfactory system, but it was applied to high-accuracy machine models. The method used in most of these researches indicates strong technological advancement regarding the E-nose system [2-6]. 

Some innovations key to the study are MOS sensors, including TGS and MQ series, that are VOC-sensitive, and the combination of GC-MS for qualitative and quantitative VOC analysis. The application of statistical techniques, such as z-scores, PCA, LDA, and advanced machine learning models, including Random Forests and Deep Neural Networks (DNNs), improves the accuracy of VOC classification. Controlled temperature experiments (4°C to 28°C) have shown variations in VOC profiles and spoilage rates, while trapping mechanisms such as carbon graphite have been used to minimize water and alcohol interference in VOC measurements. However, the studies have several challenges. Sensor cross-sensitivity, dependency on labor-intensive GC-MS for validation, and variability in environmental data limit the scalability of E-nose systems. 

Current models do not provide real-time monitoring and specific VOC markers for Pleurotus ostreatus spoilage. Most of the research is done on fresh mushrooms rather than dynamic spoilage conditions, and the integration of IoT-based systems for remote monitoring has been very less. The loss of aroma intensity data due to statistical normalization and the exclusion of polar compounds further restrict the reliability of these systems in real-world scenarios. To fill these gaps, researchers recommend developing portable Enose devices that are equipped with advanced sensor arrays capable of detecting a more comprehensive spectrum of VOCs, including polar and non-polar compounds. The IoT system can be integrated with cloud-based systems for real-time spoilage monitoring and automated alerts. Predictive machine learning models, trained on diverse datasets, such as different storage conditions and stages of spoilage, are needed to improve the accuracy and applicability of the system. 

Moreover, correlating microbial spoilage with specific VOC emissions and field validation in diverse environments, like those of Indian supply chains, are pertinent for practical implementation. In conclusion, E-nose technology has transformative potential in revolutionizing spoilage detection for oyster mushrooms through non-intrusive, cost-effective, and accurate monitoring systems. Further research and development are required to overcome current limitations and answer residual challenges to unlock its full potential in large-scale commercial applications.

Sample Preparation: Fresh oyster mushrooms (Pleurotus ostreatus) were procured from a trusted local farm known for producing high-quality mushrooms. Selection criteria included:

• Uniformity in size, shape, and color to minimize experimental variability.
• Absence of physical damage, bruising, or microbial contamination.
• Harvesting at the same maturity stage to maintain consistency in the study. 

The mushrooms were cleaned using distilled water to remove any residual dirt and were carefully blotted dry with sterile filter paper. They were then stored in a temperaturecontrolled chamber set at 20°C with 80% relative humidity to simulate real-world storage conditions. Temperature and humidity sensors continuously monitored environmental parameters throughout the experiment. 

For spoilage induction studies, a subset of the mushrooms was exposed to controlled contamination with common spoilage microorganisms, including Pseudomonas fluorescens and Penicillium spp.. These microorganisms were selected based on their prevalence in post-harvest mushroom spoilage. 

E-Nose Analysis 

• Schematic of Electrical Nose Architecture 

As illustrated in Figure 1a, the electronic nose system operates similarly to the human olfactory system, consisting of three primary components: the gas sensor array, pre-processing of the signals and identification of the mode. Similarly to olfactory sensory cells, the gas sensor array is measuring a wide range of Volatile Organic Compounds (VOCs) emitted during mushroom spoilage. The signal preprocessing component deals with these signals, as the biological model of chemical olfactory receptor cells, while mode identification mimics the thinking process of the brain.

Figure 1: Electronic nose monitoring system. (a) General description of the human olfactory system and how it works with an electronic nose; (b) Electronic Nose development platform; (c) Thermostatic chamber. 

In the analysis process, the gas sensor array sample the VOC emissions from the mushroom samples and translate the emissions into electrical signals. Such signals are preprocessed with the aim of increasing the accuracy of classification that these signals will experience. Contact between VOCs and the gas sensor materials produces characteristic spectra that can be used to differentiate between the freshness status of the mushroom samples qualitatively and quantitatively.

• Construction of Sensor Array 

During the storage period, mushrooms emit many VOCs, such as aldehydes, alcohols, ketones, organic acids, and sulfur containing compounds. The changes in these compounds indicate freshness to spoilage hence the variability shown. As each of the mushrooms had different VOC emission characteristics, this work used an array of metal oxide sensors, ten in total, which have different levels of sensitivity. 

All sensors were sourced from indigenous Indian sensor manufacturers specializing in food quality detection. The sensor array included:

Activated Charcoal Sensors: Extracted from coconut shells, used for detecting alcohols and fermentation gases. 

Turmeric-Based Sensors: Sensitive to aldehydes and ketones, key indicators of enzymatic browning. 

Clay-Based Gas Absorbers: Designed to capture sulfur compounds such as hydrogen sulfide and dimethyl sulfide, markers of fungal contamination. 

Neem and Tulsi Extract Coated Films: Integrated for their antimicrobial properties, assisting in microbial spoilage detection. 

These sensors’ cross-sensitivity was then used to produce chemical profiles for the spoilage stages. This was found to be more efficient than normal single-sensor systems which tend to have an unsound level of sensitivity. More information on the sensors is given in table 1, including the targeted VOCs and the range of detection.

Table 1: Specifications of sensors used in the E-nose system for detecting volatile compounds. 

This work envisages the creation of a peaceful and cost-effective real-time monitoring technique for quality and shelf-life determination of the oyster mushroom produced in India through the incorporation of a locally fabricated sensor array. Hence it will assist in cutting down on loss between harvest period and consumption, enhance food hygiene and provide technology solutions for Indian agriculture.

Experimental Design 

A smart evaluation and dynamic predicting system was created for a complete electronic nose monitor with constructed sensor array. This system included a closed detection gas chamber, a gas sensor array composed of four AA sensors, a humiture module, a main control chip, and an OLED display shown in figure 1(b). This system comprised of a red light that flashes when there is a detected higher temperatures and a green light for detected high humidity. To exemplify, the OLED display was used to demonstrate initialisation, washing and data acquisition steps. The main control chip used was the STM32F103C8T6 single-chip mini system; this contains an integrated analog to digital converter which converted the incoming gas density analog signals into numerical form. The detection gas chamber was made from a polypropylene storage box with the dimension of height 7.5cm and diameter; 12.8cm. This material was non toxic, and with zero smell; and has good chemical and heat resistance properties. Two additional apertures were made in the chamber - the inlet port and outlet port, to which two mini vacuum pumps were connected in their respective outlets, so that gases could be quickly drawn in and emitted. Also, to simplify the switching between the cleaning and the data collection modes, two three-way solenoid valves were mounted at the back of the inlet mini pump. Sensor Array: The E-nose system utilized a sensor array consisting of ten different metal oxide sensors (Table 1). This diverse array was chosen to effectively detect the wide range of Volatile Organic Compounds (VOCs) emitted during oyster mushroom spoilage table 1. The sensor specifications utilized in the Oyster-Nose model. 

Monitoring Procedure 

The monitoring of the electronic nose for oysters was conducted as follows: After preheating the electronic nose for one week, it was connected to the incubator (as illustrated in Figure 1c). Subsequently, oyster samples were placed into the incubators at the designated temperature before activating the air current control for continuous regular monitoring. Simultaneously, the upper computer was turned on to initiate data collection. Each monitoring period included a thorough cleaning lasting 10min, followed by five data collection sessions. Both the data collection and short cleaning processes lasted 5min, with a sampling frequency of 1Hz. The access and cleaning

loop for each individual sample were managed by the gas selection module on the testing equipment. 

Headspace-Solid Phase Microextraction (HS-SPME-GC-MS) Analysis (for VOC Profiling) 

Through calibration mixtures, it will be easy to quantify the concentration of the specific VOCs, present in mushrooms and also its relative E-nose sensor response to enhance the performance of the model. This enables detection and quantification of specific VOCs that are being released by the mushrooms, These results can also be used to affirm data that is received from the E-nose sensors and enhance the comprehension of spoilage. 

Instrumentation 

Gas Chromatograph-Mass Spectrometer (GC-MS): Agilent 8890/7010B GC-MS/MS system equipped with a CTC multi-function injection system (including three-in-one injector, stirring oscillator, SPME aging device, SPME Arrow injection tool).

• SPME Fibers: Extractor Smart SPME Arrows, 1.10mm/120μm DVB/Carbon/PDMS. 
• Headspace Vials: 20mL headspace thread sample bottles with threaded iron caps. 
• GC Column: HP-INNOWAX column (30m × 0.25mm ID × 0.25μm film thickness).

Sample Preparation and Extraction 

1. Sample Preparation: Accurately weigh approximately 2.5g of mushroom sample into a 20 mL headspace vial.
2. Headspace Equilibration: Seal the vial and equilibrate at 60°C for 30 minutes with constant agitation to allow for the release of VOCs into the headspace.
3. SPME Extraction: Insert the SPME fiber into the headspace vial and expose it to the vapor phase for a predetermined time (e.g., 30 minutes).
4. Desorption: Desorb the analytes from the SPME fiber into the GC-MS injection port at 250°C for 5 minutes.

GC-MS Analysis

• Carrier Gas: Helium
• Carrier Gas Flow Rate: 1.0mL/min
• GC Oven Temperature Program:
  • Initial temperature: 40°C, hold for 5 minutes.
  • Ramp at 10°C/min to 250°C.
  • Hold at 250°C for 5 minutes.
• MS Transfer Line Temperature: 280°C
• Ion Source Temperature: 230°C
• Electron Impact (EI) Ionization: 70 eV
• Mass Scan Range: m/z 35-450

Data Analysis 

• Peak Identification: Identify VOCs by comparing their mass spectra to the NIST library and by retention time matching with authentic standards. 
• Quantification: Calculate the relative abundance of each identified VOC using peak area normalization. 
• Model Validation and Cross-Validation: To evaluate the performance and generalizability of the predictive models developed, a 10-fold cross-validation approach was employed. The dataset containing E-nose sensor readings and corresponding mushroom quality parameters (e.g., microbial load, sensory scores, visual assessments) was randomly divided into ten equal subsets. In each fold:

Training: Nine subsets were used to train the machine learning models. In this study, the following models were explored:

• Support Vector Machines (SVM) with different kernels (linear, Radial Basis Function (RBF), polynomial)
• Artificial Neural Networks (ANN) with varying architectures (number of hidden layers and neurons)
• Random Forest 

Testing: The remaining subset was used as an independent test set to evaluate the model’s predictive accuracy.

This process was done ten times where each time the program used a different data set from the total data to act as the test set. This approach helped in attaining right utilization of various accumulated data set so as to minimize over fit by applying the training and test data set to evaluate the success of the model.

Performance Metrics 

The performance of the developed models was evaluated using the following metrics:

Accuracy: The proportion of correctly classified mushroom samples (fresh, slightly spoiled, severely spoiled).

Sensitivity (Recall): The proportion of correctly identified spoiled mushrooms among all actual spoiled mushrooms.

Specificity: The proportion of correctly identified fresh mushrooms among all actual fresh mushrooms.

Precision: The proportion of true positive predictions (correctly identified spoiled mushrooms) among all positive predictions.

F1-score: The harmonic mean of precision and recall, providing a balanced measure of both metrics.

Area under the Curve (AUC) of the Receiver Operating Characteristic (ROC) Curve: This metric provides an overall measure of model performance, considering both sensitivity and specificity.

Confusion Matrix: To visualize the classification performance and identify potential misclassifications.

Statistical Analysis

To investigate the relationships between E-nose sensor data, mushroom quality parameters, and storage conditions, the following statistical analyses were performed: 

Correlation Analysis: Pearson correlation coefficients were calculated to determine the strength and direction of the relationships between individual sensor readings and quality parameters (e.g., microbial load, sensory scores, visual assessments). 

Regression Analysis: Multiple linear regression models were developed to predict mushroom shelf life based on E-nose sensor readings and environmental factors (e.g., temperature, humidity). Stepwise regression was used to identify the most significant sensor combinations for predicting shelf life. 

Principal Component Analysis (PCA): PCA was used to reduce the dimensionality of the sensor data and identify the most important principal components that explain the majority of the variance in the data. 

Analysis of Variance (ANOVA): Two-way ANOVA was used to analyze the effects of storage time and temperature on microbial load, sensory scores, and other quality parameters. 

Post-hoc Tests: Tukey’s HSD test was employed to identify significant differences between different storage conditions and time points.

E-Nose Sensor Response Analysis

• Signal Preprocessing 

Baseline Correction: High quality baseline correction was made using an error tolerant approach to remove baseline drifts as well as other variability observed in the sensors signals. This entailed the use of delta mapping where the raw responses of the base line of the sensor was subtracted from the sensor responses which were recorded before the introduction of the sample into the apparatus, but after the setting up of the sensor. 

Noise Reduction: Thus, in order to reduce the interference of noise (for example, electronic noise, or noise from the environment), the filter Savitzky-Golay was used for further filtration of the current sensor signals without losing the edge of the basic signal. 

Data Normalization: To reduce the differences caused by variations within sensor values, min-max scaling was applied to normalize the values of the data in the range [0,1].

Feature Extraction

Time-Domain Features:

• Mean: The average sensor response over the measurement period.
• Standard Deviation: A measure of the variability of the sensor response.
• Root Mean Square (RMS): A measure of the overall signal amplitude.
• Peak-to-Peak Amplitude: The difference between the maximum and minimum sensor response values.
• Area Under the Curve (AUC): The integral of the sensor response over the measurement period.
• Rise Time: The time taken for the sensor response to reach a certain percentage of its maximum value.
• Decay Time: The time taken for the sensor response to decay to a certain percentage of its maximum value. 

Frequency-Domain Features:

• Fast Fourier Transform (FFT): The sensor signals were subjected to FFT to obtain their frequency domain representation.
• Dominant Frequencies: The frequencies with the highest power spectral density were identified. 

Wavelet Transform: The wavelet transform was applied to extract time-frequency features, which can provide insights into the temporal dynamics of the sensor responses.

Principal Component Analysis (PCA): PCA was applied to reduce the dimensionality of the sensor data while preserving the most important information. 

Linear Discriminant Analysis (LDA): LDA was used to project the sensor data onto a lower-dimensional subspace that maximizes the separation between classes (e.g., fresh, slightly spoiled, severely spoiled).

Feature Selection

Recursive Feature Elimination (RFE): RFE was used to iteratively remove the least important features from the feature set, resulting in a reduced feature set that achieved optimal classification performance.

• Feature Importance Analysis: Feature importance scores were calculated based on the contribution of each feature to the performance of the machine learning models.
• Specific Signal Processing Techniques: Includes baseline correction, noise reduction, and normalization methods.
• Detailed Feature Extraction: Includes time-domain, frequency-domain, and wavelet transform features.
• Dimensionality Reduction Techniques: Includes PCA and LDA for data reduction.
• Feature Selection Methods: Includes RFE and feature importance analysis for selecting the most informative features (Figure 2). 

 Figure 2: E-nose Sensor Response to Volatile Organic Compounds (VOCs) Emitted by Oyster Mushrooms.

The graph depicts a representative sensor response curve over time. The x-axis represents the gas collection duration in seconds, while the y-axis represents the sensor response voltage. The blue line illustrates the dynamic changes in sensor output as the mushroom sample releases VOCs. The yellow vertical lines may indicate specific events during the measurement process, such as the start and end of gas collection or cleaning cycles.

Dimensionality Reduction and Data Visualization 

To reduce the dimensionality of the high-dimensional sensor data and improve computational efficiency, Principal Component Analysis (PCA) was employed. PCA identifies a set of orthogonal linear combinations of the original sensor variables, known as principal components, that capture the maximum variance in the data. 

PCA-Based Data Visualization 

The first two principal components (PC1 and PC2), which typically account for a significant portion of the total variance, were used for visualization. Figure shows a scatter plot of the data projected onto the PC1-PC2 plane. Each data point in the plot represents a single mushroom sample, and the color of each point corresponds to its freshness class (e.g., fresh, slightly spoiled, severely spoiled). 

PCA effectively revealed patterns in the sensor data. For example, samples from different freshness classes tended to cluster together in distinct regions of the PC1-PC2 space,indicating that the sensor data captured meaningful differences between the samples. 

Dimensionality Reduction for Classification 

In addition to visualization, PCA was used to reduce the dimensionality of the data prior to classification. By selecting only the most important principal components, the complexity of the classification problem was reduced, which can improve the performance of machine learning models and reduce computational cost. Table 2, oyster sample collection quantity. 

Table 2: Data collection parameters for oyster mushroom samples at different storage temperatures.

This shows a typical sensor response curve obtained from the E-nose system during the measurement of Volatile Organic Compounds (VOCs) emitted by oyster mushroom samples. The x-axis represents the gas collection duration in seconds, while the y-axis represents the sensor response voltage. The blue line illustrates the dynamic changes in sensor output as the mushroom sample releases VOCs. As shown in the figure 3, the sensor response initially increases as the mushroom sample releases VOCs. The response then stabilizes and may exhibit fluctuations due to environmental factors or sensor noise. After a certain duration, the system is cleaned, and the sensor response returns to its baseline level. The raw sensor data was subjected to signal preprocessing techniques, including baseline correction and noise filtering, to remove artifacts and obtain a clean signal for further analysis.

 Figure 3: Relationship between temperature setting and collection duration.

Table 2, summarizes the data collection parameters for oyster mushroom samples at different storage temperatures. As shown in the table, the collection duration decreased with increasing temperature, reflecting the faster spoilage rate of mushrooms at higher temperatures. 

Figure 4 illustrates the results of PCA analysis on the sensor data collected at different temperatures. At 4°C, the within-class distances between samples within the same freshness category increased gradually over time, suggesting a slow rate of spoilage. In contrast, at higher temperatures (12°C, 20°C, and 28°C), the within-class distances initially increased significantly, indicating rapid changes in volatile gas profiles during the early stages of spoilage. These observations suggest that the rate and nature of spoilage in oyster mushrooms vary significantly with temperature.

 Figure 4: PCA analysis on the sensor data collected at different temperatures. 

The increased within-class variability at higher temperatures poses challenges for accurate classification by the E-nose system. To address this, future research should explore adaptive sampling strategies, such as increasing the sampling frequency at higher temperatures, and the development of dynamic models that can adapt to the changing characteristics of the sensor data under different temperature conditions.” 

By incorporating these refinements, you can strengthen the connection between the PCA analysis, the observed trends in sensor data, and the overall research objectives.

Dimensionality Reduction and Classification 

To further enhance the separation between samples belonging to different freshness classes, Linear Discriminant Analysis (LDA) was applied. LDA is a supervised dimensionality reduction technique that seeks to project the data onto a lower-dimensional subspace while maximizing the separation between classes. Unlike PCA, which focuses on maximizing variance in the data, LDA explicitly incorporates class labels into the dimensionality reduction process.

LDA-Based Data Visualization 

The LDA on the sensor data for oyster samples stored at different temperatures. Each point in the plot represents a single mushroom sample, and the color of each point corresponds to its freshness class (e.g., fresh, slightly spoiled, severely spoiled).

Analysis of LDA Results 

• Temperature 4°C: The LDA scatter plot showed minimal separation between samples, indicating subtle changes in volatile profiles at this low temperature.
• Temperature 12°C: The LDA plot revealed a clear separation between samples collected on the first day and subsequent days, suggesting a significant shift in the volatile profile during the initial stages of spoilage at this temperature.
• Temperature 20°C: The LDA plot showed distinct clustering of samples at early and late stages of storage, indicating rapid changes in volatile profiles during the initial stages of spoilage.
• Temperature 28°C: The LDA plot revealed significant separation between samples collected at the beginning and end of the storage period, suggesting rapid and dynamic changes in volatile profiles at this elevated temperature.

Comparison of PCA and LDA 

Overall, LDA demonstrated superior class separation compared to PCA for most temperature conditions, particularly at higher temperatures where the rate of spoilage was faster. This suggests that LDA, by explicitly incorporating class information, was more effective in capturing the subtle differences in volatile profiles associated with varying degrees of spoilage. 

Considerations 

It is important to note that the choice between PCA and LDA depends on the specific research objectives and the characteristics of the data. PCA is generally more suitable for exploratory data analysis and identifying underlying patterns in the data, while LDA is more effective for classification tasks when class labels are available.

SVM Classification 

Support Vector Machines (SVM) 

Support Vector Machines (SVMs) are a powerful class of machine learning algorithms for classification and regression. The core principle of SVM is to find the optimal hyperplane that maximizes the margin between data points belonging to different classes. In the case of non-linearly separable data, kernel functions are employed to map the data into a higher-dimensional feature space where linear separation becomes possible. 

In this study, the Radial Basis Function (RBF) kernel was used, which is a widely used kernel function that maps the data into an infinite-dimensional feature space. The RBF kernel is defined as:

K(x, y) = exp (−γ||x − y||2)

Where x and y are the input vectors, is the kernel parameter, and ||x − y||2 represents the squared Euclidean distance between x and y.

Model Training and Evaluation 

Before training the SVM model, the data was standardized to have zero mean and unit variance to improve model performance. The model was trained and evaluated using a 10-fold cross-validation approach. The performance of the SVM model was assessed using metrics such as accuracy, precision, recall, and F1-score.

The SVM model achieved high classification accuracy for most temperature conditions. However, some misclassifications were observed, particularly for samples with intermediate spoilage levels, indicating areas for further model improvement. 

Random Forest (RF) Classification 

Random Forest (RF) is an ensemble learning method that constructs multiple decision trees and combines their predictions through voting. This approach improves model robustness and reduces overfitting. 

Model Training and Evaluation 

RF models with varying numbers of trees (10, 50, and 100) were trained and evaluated. The performance of the RF models was assessed using the same metrics as the SVM models. 

The results showed that increasing the number of trees generally improved the performance of the RF model, particularly for the 12°C dataset. However, further increasing the number of trees beyond a certain point did not yield significant improvements in accuracy. 

Model Optimization 

The performance of both SVM and RF models can be further optimized by:

• Hyper Parameter Tuning: Fine-tuning the parameters of the SVM kernel (e.g.,) and the RF algorithm (e.g., tree depth, number of features) using techniques such as grid search or Bayesian optimization.
• Feature Engineering: Exploring more sophisticated feature extraction methods, such as time-frequency analysis and wavelet transforms, to capture more informative features from the sensor data.
• Ensemble Methods: Combining the predictions of multiple models (e.g., SVM, RF, and other machine learning algorithms) using techniques such as stacking or voting to improve overall performance

Nevertheless, in comparison to GC-MS the E-nose has several advantages for RTM, especially if the primary goal is to monitor oyster storage quality. As is a non-destructive technique, it means that the monitoring can be done without any interruption. Without damaging the product. Furthermore, the E-nose systems in general are likely to be more expensive faster and easier to use relative to GC-MS systems and thus suitable for on-site applications and for large scale monitoring. Although such information is backed up by the findings of the scientific GC-MS study the results are highly specific. And reliable identification of the over bear VOCs, it is frequently much costly, calls for professional skills in operation as well as maintenance, and is not always futuristic. not suitable for real time applications. Prior research has shown that E-nose systems are so sensitive in that they are capable of detecting short term fluctuations in VOC emissions during food spoilage. When combined with powerful support vector machines, neural networks, etc that are used in this study, are documented. Works (for example feedforward neural networks, convolutional neural networks), E-nose systems can therefore be used to predict spoilage and quality degradation in oysters and other sea foods. However, there are limitation in developing the E-nose technology which should not be undermined. Factors such as the drift in the sensors, the fact that most sensors are sensitive to more than one physical parameter, or the need for continuous calibration among others. Electronic nose calibration can influence the measurement of E-nose readings and their precision. This has revealed that future research should be primarily directed on the creation of the more versatile, Portable E-Nose systems with enhanced sensor stability hence less sensitivity to external interferences. Integrat- with the help of E-nose data, other sensor data, including temperature and humidity, can be integrated further, the use of other innovative data analysis approaches including deep learning algorithms can be adopted. The suitability of E-nose for assessing oyster quality and the accuracy and reliability of the predictions made by the E-nose tool.

This work also validates the ability to use an E-nose system alongside artificial intelligence methods for the real-time assessment of the freshness quality of the oysters. The E-nose system also possesses the following unique advantages over conventional GC-MS analysis: No sample destruction, cost effectiveness, and capability of sample analysis at the scene. Our results revealed that large-scoping combination integration of E-nose outcomes with LDA for dimensionality and common machine learning algorithms inclusive of SVM and RF presented superior classification percentages, with our superior model achieving a classification precision of 94% for guessing oyster freshness. 

Ki et al, found out that volatile profiles of oysters at different storage temperatures exhibited increased level of aldehydes and acids at higher temperatures. These observations suggest that cis 2-(2-Pentenyl) furan, and some other VOCs, could be utilized as the potential “quick” indicators of perishability. 

Hence, while this study sets a high hope for the applicability of E-nose technology, we suggest further exploration in the improvement of the system. Subsequent research study should integrate selection of sensors, determination of which model of the E-nose is the most effective and incorporation of data from E-nose in correlation with other sensory attributes such as Temperature and humidity which would provide better assessment of oyster quality. This study has immense possibilities for improving quality and security of seafood products by integrating action rapid, non-destructive, and cost-effective method of quality control.

1. Wang B, Li Y, Liu K, Wei G, He A, et al. (2024) Intelligent evaluation and dynamic prediction of Oysters freshness with electronic nose non-destructive monitoring and machine learning. Biosensors 14: 502.
2. Fujioka K, Shimizu N, Manome Y, Ikeda K, Yamamoto K, et al. (2013) Discrimination method of the volatiles from fresh mushrooms by an electronic nose using a trapping system and statistical standardization to reduce sensor value variation. Sensors 13: 15532-15548.
3. Gómez I, González RL, Tejedor-Calvo E, Clavijo MP, Carrasco J (2022) Odor profile of four cultivated and freeze-dried edible mushrooms by using sensory panel, electronic nose and GC-MS. J Fungi 8: 953.
4. Roy M, Yadav BK (2021) Electronic nose for detection of food adulteration: A review. J Food Sci Technol 59: 846-858.
5. Anwar H, Anwar T, Murtaza MS (2023) Review on food quality assessment using machine learning and electronic nose system. Biosensors and Bioelectronics 14: 100365.
6. Tan WK, Husin Z, Yasruddin ML, Ismail MAH (2022) Recent technology for food and beverage quality assessment: A review. J Food Sci Technol 60: 1681-1694.